To contribute to the best practice discussion, I recently engaged in an exercise to compare open data sites based on their functionality and data quality and realised that they could be grouped into a spectrum by 'generation', with distinct generational levels from a basic data index with variable data format and quality up to a highly sophisticated cross-jurisdictional platform embedding visualisation, error correction and community data sets.
I've formalised this thinking into five generations and published it as 'Mapping open data site generations' in my blog: http://egovau.blogspot.com.au/2012/09/mapping-open-data-site-generations.html
It could be useful to jurisdictions asking the question 'what next for our open data site?' or who are considering setting one up and want to deliver a good solution.
Feel free to reuse and share under the Creative Commons license :)
Cheers,
Craig